Amplification of spatially isolated adenosine pathway by tumor–macrophage interaction induces anti-PD1 resistance in hepatocellular carcinoma

Tumor tissue samples and/or adjacent nontumorous samples were obtained from HCC patients who underwent curative liver resection between 2018 and 2019 at the Liver Cancer Institute, Zhongshan Hospital, Fudan University. Biopsy samples and blood samples were collected from advanced HCC patients before nivolumab therapy. Tissue microarray (TMA) was constructed by Shanghai Biochip Co. Ltd. (Shanghai, China). All the HCC tissues were reviewed and confirmed histologically by H&E staining. Detailed materials are listed in Additional file 2. The collection of human samples was approved by the Zhongshan Hospital Research Ethics Committee. Written informed consent was obtained from each patient.

All the identified cell lines were provided from Key Laboratory of Carcinogenesis and Cancer Invasion, Ministry of Education of P.R.C or National Collection of Authenticated Cell Cultures, P.R.C. Cell lines HepG2, HCCLM3, MHC97H, Huh-7, PLC/PRF/5, H22, and Hepa1-6 were cultured in DMEM with 10% fetal bovine serum. The cell lines THP-1 and Li-7 were cultured in RPMI-1640 with 10% fetal bovine serum. All the cells were cultured at 37 °C in a 5% CO₂ incubator.

qPCR and Western blot analysis were performed as our previous study described [20]. The antibodies, primers, and detailed materials are listed in Additional file 2.

RNA ISH, RNA FISH, RNA pulldown, and RNA sequencing were performed as our previous study described [19, 20]. The detailed methods and materials are listed in Additional file 2.

Male C57BL/6 J mice aged 6 weeks were purchased from Vital River Laboratory Animal Co., Ltd. (Beijing, China). C57BL/6-Entpd1^fl/fl was constructed by Cyagen (Suzhou, Jiangsu, China). C57BL/6-Lyz2^CreERT2 were constructed by Shanghai Model Organisms Center, Inc. (Shanghai, China). C57BL/6-Lyz2^CreERT2 × C57BL/6-Entpd1^fl/fl mice were constructed in our laboratory. Silencing ENTPD1 in macrophage was activated by treating with tamoxifen (dissolved in corn oil, 20 mg/ml by shaking overnight at 37 °C) i.p. (100ul/mouse, for 5 days).

Subcutaneous tumor xenograft in mice model: 1 × 10⁶ H22 mice liver tumor cells in 0.1 ml of DMEM were implanted subcutaneously in the right flank of the mice. The tumor volume was screened by caliper every 3 days.

Orthotopic tumor xenograft in mice model: 1 × 10⁶ H22 mice liver tumor cells labeled luciferase (H22-luc) in 50uL Matrigel were implanted to construct orthotopic liver xenografts in the mouse. The tumors were screened by optical in vivo Imaging system.

Treatment was performed 7 days later after tumor construction. Anti-mouse PD1 (HRP00262-012, 0.85% saline vehicle, provided as gifts by Hengrui Medicine Com., Jiangsu, China) was intraperitoneally injected (10 mg/kg, 3 times every week, for 2 weeks). POM1 was intraperitoneally injected (5 mg/kg, 4 times every week, for 2 weeks). Clophosome was intraperitoneally injected (0.2 μl/mouse, every week, for 2 weeks).

Optical in vivo imaging is detailed in Additional file 2.

In brief, exosomes were pelleted by ultracentrifugation and used for the following experiments. Tumor dissociation was performed following the protocol from Tumor Dissociation Kit, mouse (#130–096-730, Miltenyi Biotec) or Tumor Dissociation Kit, human (#130–095-929, Miltenyi Biotec). All the detailed methods are listed in Additional file 2.

Isolated single-cell suspensions from the blood or tissue of human or mouse were centrifuged (350 g, 5 min, at 4 °C) in a centrifuge and resuspended in staining buffer (1% FBS in PBS) on ice. Fc Blocks (1 μl for each tube) were used before staining.

For surface staining, the antibodies were incubated with cell suspensions for 20 min in dark place at 4 °C. For intracellular staining, saponin was used after fixation by 4% paraformaldehyde, and the antibodies were incubated with cell suspensions for 20 min in dark place at 4 °C. For nuclear staining, after fixation and permeabilization by buffer set, the antibodies were incubated with cell suspensions for 20 min in dark place at 4 °C.

For staining of peripheral blood samples, after staining, red blood cells were lysed for 4 min at room temperature in red blood cell lysing buffer. Cell suspensions were washed with PBS (4 °C) for two times and filtered using a strainer (70 µm) before flow cytometry analysis. Data were analyzed by FlowJo 10.6.2 software.

The surface staining for sorting was the same as that for flow cytometry analysis, except that all the procedure was operated in a sterile environment.

All the antibodies and related reagents are listed in Additional file 2.

Other methods are detailed in Additional file 2.

To investigate the role of circRNAs in HCC patients treated with anti-PD1, 6 patients with advanced HCC treated with nivolumab (human anti-PD1 antibody, Bristol Myers Squibb) were analyzed based on retrospective data. Of these patients, 3 were evaluated as ‘partial response (PR)’ by iRECIST[21] and defined as having anti-PD1-sensitive HCCs, while the others showed ‘progressive disease (PD)’ and were defined as having anti-PD1-resistant HCCs (Fig. 1a). Tumor samples from these 6 patients were collected by puncture biopsy before immunotherapy and were used to perform RNA sequencing. We found that 24 circRNAs were upregulated in anti-PD1-resistant HCCs, and 20 circRNAs were downregulated. Among these differentiated circRNAs, hsa_circ_0001663 (circTMEM181) consistently showed the greatest difference between anti-PD1-resistant HCCs and anti-PD1-sensitive HCCs (Fig. 1b, Additional file 1: Fig. S1a). circTMEM181 splices were generated from exons 5, 6, and 7 of TMEM181 (Fig. 1c). We used a divergent circTMEM181 primer pair to perform polymerase chain reaction (PCR) and amplified the back-spliced products (168 bp) (Additional file 1: Fig. S1b). Head-to-tail splicing in the PCR products was confirmed by Sanger sequencing (Fig. 1c). Then, 60 pairs of HCC and adjacent tissues were used for real-time quantitative PCR (qPCR), showing that circTMEM181 expression in tumor tissues was significantly higher than that in para-tumor tissues (Fig. 1d). This was confirmed by in situ hybridization in a tissue microarray (TMA) including 204 HCC patients (Fig. 1e and f). These 204 patients were grouped into high and low cricTMEM181 expression according to the median. Survival analysis showed that patients with high cricTMEM181 expression had a shorter OS than those with low expression (median survival: 40 ± 7.4 vs. 70 ± 18.8 months). Moreover, patients with high cricTMEM181 expression also tended to have early recurrence after operation (median disease-free survival time: 19 ± 2.8 vs. 35 ± 9.7 months) (Fig. 1g and h). Interestingly, correlation analysis showed that patients with high circTMEM181 expression had a significantly high rate of microvascular invasion (p = 0.01) (Fig. 1i, red frame). Meanwhile, univariate analysis showed that circTMEM181 (high expression), larger tumors (≥ 5 cm), and microvascular invasion were risk factors for OS of HCC patients (Fig. 1i). In multivariate Cox proportional hazards mode, circTMEM181 (high expression), larger tumors (≥ 5 cm), and microvascular invasion were independent prognostic indicators for OS (Fig. 1i). These results indicate that elevated circTMEM181 plays an important role in HCC progression.

circTMEM181 expression was analyzed in six human HCC cell lines. Among these cells, Huh-7 had the highest circTMEM181 expression, while HepG2 had the lowest level (Fig. 2a). Human HCC cell lines and mouse HCC cell line H22 stably overexpressing or with knocked down circTMEM181 were successfully constructed (Additional file 1: Fig. S1C). Unexpectedly, circTMEM181 expression in HCC cell lines did not influence their metastatic potential or proliferative capacity (Fig. 2b and c). Meanwhile, we also found no statistical difference in the proliferative capacity between H22 cell lines overexpressing circTMEM181 (H22^OE) and the control (H22^Ctrl) (Additional file 1: Fig. S2). Next, we investigated the role of circTMEM181 expression in HCC in vivo. H22^OE or H22^ctrl that were previously labeled with luciferase were used to construct orthotopic liver xenografts in C57 mice. We found that tumors in H22^OE + IgG group grow faster than that in H22^ctrl + IgG group, indicating that circTMEM181 may promote tumor progression in vivo (Additional file 1: Fig. S3). Interestingly, consistent with our clinical discovery cohort with anti-PD1 therapy (Fig. 1a), H22^ctrl was sensitive to anti-PD1 antibody, while H22^OE showed a negative response to anti-PD1 antibody therapy (p < 0.01) (Fig. 2d). In this orthotopic tumor model with anti-PD1 therapy, OS in the H22^ctrl group was also significantly longer than that in the H22^OE group (p = 0.033) (Fig. 2e). There were significantly more metastatic lung nodules in the H22^OE group than in the H22^ctrl group (Fig. 2f, right). The sizeable difference between circTMEM181 function in vivo and in vitro indicated that circTMEM181 might mainly affect the immune microenvironment of HCCs rather than tumor cells. Thus, we harvested tumors from the two mouse models that underwent anti-PD1 therapy and analyzed the immune cell composition in CD45⁺ tumor-infiltrating leukocytes (TIL) by flow cytometry. Eight clusters were identified: dendritic cells (DC, CD11b⁺ CD11c⁺), M2 (CD11b⁺ F4/80⁺ CD163⁺), other macrophages (other Mphi, CD11b⁺ F4/80⁺ CD163⁻), CD8⁺ T (CD11b⁻ CD8⁺ CD4⁻), CD4⁺ T (CD11b⁻ CD4⁺ CD8⁺), NK (CD11b⁻ NK1.1⁺), B cells (CD11b⁻ CD19⁺), and CD11b⁺ myeloid cells other than DC or macrophages (Fig. 2g Top). As expected, the H22^OE group showed a distinct immune cell profile in the tumor microenvironment compared with the H22^ctrl group after anti-PD1 therapy (Fig. 2g Bottom). Particularly, a decreased CD8⁺ T cell population and an increase in F4/80⁺ CD163⁺ M2 macrophages were found in the tumor microenvironment of the H22^OE group (Fig. 2g and h). We also investigated CD206 expression (another classical marker for M2 macrophage in mice) by flow cytometry and found an increase in CD206⁺ M2 macrophages in H22^OE + αPD1 group (Additional file 1: Fig. S4), consistent with the results that we found in CD163⁺ M2 macrophages. By TMA, we further confirmed that HCC patients with high circTMEM181 expression (circTMEM181^high) had increased M2 macrophage infiltration (CD163⁺) and decreased CD8⁺ T cell infiltration compared to patients with lower circTMEM181 expression. However, no statistically significant differences in the number of total macrophages (CD68⁺), CD4⁺ T cells, NK cells (CD56⁺), or B cells (CD19⁺) were found (Fig. 2i and j). Our results indicate that overexpression of circTMEM181 promotes HCC development by reshaping the inhibitory immune microenvironment in HCC.

To further elucidate the mechanism of influence of HCC cell-derived circTMEM181 in tumor-infiltrating immune cells, CD8⁺ T cells isolated from human peripheral blood mononuclear cells (PBMCs) were co-cultured with HCC cells (Fig. 3a). When co-cultured with Huh-7 cells overexpressing circTMEM181 (Huh-7^OE) for 72 h, CD8⁺ T lymphocytes showed a slight but statistically inhibited proliferation ability compared to the control (Fig. 3b). Given that macrophages were significantly influenced by circTMEM181 expression (Fig. 2g-j), Huh-7 cells were co-cultured with THP-1 macrophages for 72 h, and then the supernatant was collected and incubated with CD8⁺ T cells. Interestingly, CD8⁺ T lymphocytes cultured with the THP-1 and Huh-7^OE co-culture supernatant (Sup.) showed obviously inhibited proliferation ability compared to those co-cultured with Huh-7^OE or THP-1 (Fig. 3b). Immune checkpoint molecule PD1 and T cell immunoglobulin mucin family member 3 (TIM3) expression was upregulated on CD8⁺ T cells after co-culturing with THP-1 and Huh-7^OE supernatant, but we found no different expression of PD1 or TIM3 in CD8⁺ T cells between co-culturing with Huh-7^OE and co-culturing with Huh-7^Ctrl (Fig. 3c). These results reveal that overexpressing circTMEM181 in HCC cells can affect immune cell status in vitro. HCC cells might interfere with the proliferation of CD8⁺ T cell and induce their exhaustion by interacting with macrophages.

Previous studies denoted the importance of exosomes in the interrelationship between tumor and immune cells [22, 23]. Given that the in vitro co-culture system separated tumor cells from PBMCs in independent layers (Fig. 3a), we ruled out the possibility that tumors overexpressing circTMEM181 directly influenced immune cells. Thus, secretory exosomes in media from various HCC cell lines were enriched using ultracentrifugation. The structure and size distribution of enriched exosomes were analyzed by transmission electron microscopy and nanoparticle tracking analysis (Fig. 3d). Exosomal markers CD63 and TSG101 were also detected on enriched exosomes from human and mouse HCC cell lines (Fig. 3e). The circTMEM181 expression level in enriched exosomes was positively related to circTMEM181 expression in corresponding HCC cells (Fig. 3f). Similarly, circTMEM181 expression in the exosome was upregulated when circTMEM181 was overexpressed in HCC cells (Fig. 3f). Further, THP-1 cells expressed slight circTMEM181, which was significantly enhanced when they were co-cultured with Huh-7^ctrl or Huh-7^OE (Fig. 3g). In addition, circTMEM181 expression in THP-1 cells was largely decreased upon addition of exosome generation inhibitor GW4869 (Fig. 3g). Next, we co-cultured THP-1 cells with exosomes from Huh-7^OE pre-labeled with PKH-67. An immunofluorescence assay confirmed that THP-1 cells could internalize exosomes from Huh-7^OE (Fig. 3h). This indicates that HCC cells secrete exosomal circTMEM181 and influence macrophage function.

Sponging and decoying miRNAs is a major mechanism by which circRNAs regulate cell functions [14]. Here, we performed RNA immunoprecipitation (RIP) to identify miRNAs that directly interact with circTMEM181. Biotin-circTMEM181-specific probes were transfected into THP-1 cells (THP-1^circOE), using transfection of biotin alone as a control (THP-1^ctrl), and RNA sequencing was performed followed by RIP (Fig. 3i). More than tenfold enriched miRNAs were identified, and miR-488-3p and miR-1298 had the top two scores and were selected for further verification (Fig. 3i and j).

Wild-type circTMEM181 (circTMEM181^WT) or mutant circTMEM181 (circTMEM181^MU) without miR-488-3p or miR-1298 binding sites were cloned into pLG3 vectors (Fig. 3j). Luciferase reporter activity was then examined with vector-transfected 293 T cells. Our results showed that miR-488-3p and miR-1298 mimic reduced luciferase activity in the circTMEM181^WT group but not in the circTMEM181^MU group (Fig. 3k). Interestingly, miR-488-3p exerted even less luciferase activity in the circTMEM181^WT group than miR-1298. Additionally, biotinylated miR-488-3p could pull down much more circTMEM181 than biotinylated miR-1298 (Fig. 3l), indicating that miR-488-3p has a direct interaction with circTMEM181.

Flow cytometry analysis of CD8⁺ T lymphocytes co-cultured with THP-1 and Huh-7^OE showed that the inhibited proliferation ability of CD8⁺ T lymphocytes could be rescued by manipulating miR-488-3p in THP-1 (Fig. 3m). Further, a fluorescent in situ hybridization (FISH) assay showed circTMEM181 colocalized with miR-488-3p in the cytoplasm of THP-1^circOE (Fig. 3n).

To detail the function and mechanism of circTMEM181 in macrophages, we performed stable isotope labeling by amino acids in cell culture (SILAC) to compare the protein profiles of THP-1 cells co-cultured with exosomes from HepG2^OE (THP-1^co−exoOE) or HepG2^Ctrl (THP-1^co−exoCtrl) (Fig. 4A). In total, 165 upregulated proteins were found in THP-1^co−exoOE (|Log₂(H/L)|> 1), compared to THP-1^co−exoCtrl (Fig. 4b). KEGG and GO analysis showed that the upregulated proteins were majorly involved in immune-associated pathways such as neutrophil activation involved in immune response, myeloid cell activation involved in immune response, and the adaptive immune system (Fig. 4b). Additionally, we used biotinylated miR-488-3p to immunoprecipitate mRNAs from THP-1^co−exoOE and identified 588 mRNAs as potential targets of miR-488-3p (Fig. 4c). By overlapping the potential targets with the upregulated proteins, 12 genes were identified as potential molecules downstream of the circTMEM181-miR-488-3p axis (Fig. 4c). Among them, CD39 (ectonucleoside triphosphate diphosphohydrolase 1, ENTPD1) and PLAC8 (placenta specific 8) had more than two positions paired with miR-488-3p (Fig. 4d). ENTPD1, with a high fold change in THP-1^co−exoOE/THP-1^co−exoCtrl, was chosen for further study because it is involved in immune-associated function (Fig. 4a).

We found an enhanced level of ENTPD1 mRNA in THP-1 after forced circTMEM181 (THP-1^circOE) expression (Fig. 4e), verifying the relationship between circTMEM181 and CD39 in macrophages. THP-1 also expressed a high level of CD39 when cultured with exosomes from HepG2^OE (co-HepG2^OE Exo.), but the upregulation of CD39 could be rescued by administering GW4869 (Fig. 4e). Similarly, the level of CD39 in THP-1 was upregulated after culturing with exosomes from Huh7^Ctrl (co-Huh-7^Ctrl Exo.) and could be reversed after supplementing with simiR-488-3p (co-Huh-7^Sh Exo. + simiR) (Fig. 4e and f).

CD39 is a key enzyme in initiating activation of the ATP–adenosine pathway. By hydrolyzing extracellular ATP (eATP) and ADP to AMP, CD39 can weaken the signal of the immune reaction stimulated by eATP in the tumor microenvironment. Analysis of 50 paired HCC and corresponding normal tissue from TCGA showed that CD39 was overexpressed in HCC tissues (Additional file 1: Fig. S6A). HCC patients with high CD39 expression also showed a poorer OS (Additional file 1: Fig. S6B). We also analyzed scRNA-seq data from three GEO datasets (GSE140228 (10X), GSE140228 (Smartseq2) and GSE125449) and found that tumor-infiltrating macrophages expressed a high degree of CD39 (ENTPD1) among all immune cell types (Fig. 4g and h).

We next investigated the function of CD39 in HCC. Recent studies have observed a high level of CD39 in tumor-infiltrating immune cells from many tumors [12, 24, 25]. Some tumors including kidney, lung, testicular, and thyroid tumor cells, but not HCC, express a high level of CD39 [26]. Here, we found that THP-1 macrophages, but not HCC cells, expressed a high level of CD39 (Fig. 5 A, Additional file 1: Fig. S6C and D). Immunohistochemistry (IHC) analysis of TMAs showed CD39-positive expression on the cell membrane and cytoplasm of immune cells, vascular endothelial cells, and fibroblasts, but only some tumor cells in tumor tissues (Additional file 1: Fig. S6E). Consistent with the results from TCGA, our data also showed that higher tumor CD39 in HCC patients was associated with poorer prognosis in terms of shorter OS and earlier recurrence (Additional file 1: Fig. S6F and G).

Extracellular ATP (eATP) is degraded into adenosine (ADO) through an ATP–adenosine pathway. Within the tumor microenvironment, accumulated eATP could promote immune activity. However, upregulated ADO has been found to be related to the tumor immunosuppressive microenvironment, and CD39/CD73 are the two key enzymes mediating the ATP–ADO process. CD39 can degrade eATP into ADP and AMP, and then CD73 degrades AMP into ADO. Interestingly, most HCC cell lines expressed high CD73, and Li-7 expressed the highest level of CD73 among them (Fig. 5a). Notably, CD73 expression on HCC cell lines was in a circTMEM181-independent manner (Fig. 5a). TMA analysis also showed that CD73 expression was ubiquitous in HCC cells (Additional file 1: Fig. S7A). Specially, CD73 was not detected on THP-1^circOE cells (Fig. 5a). To address the relationship among CD39 expression in macrophages, CD73 expression in HCC cells, and circTMEM181 expression, we co-cultured THP-1 and TAM (tumor-associated macrophages) isolated from HCC tissues with Li-7 overexpressing circTMEM181 (Li-7^OE). Flow cytometry analysis showed that CD39 expression in THP-1 or TAM was enhanced after co-culturing with Li-7^OE, but could be blocked by GW4869 (Fig. 5b). Meanwhile, CD73 was detected on Li-7^OE, but not on TAM or THP-1, even upon culturing with HCC overexpressing circTMEM181 (Fig. 5B). Public scRNA-seq data supported our results that macrophages in HCC express a high level of CD39 but a low level of CD73 (Fig. 4h). To eliminate the potential effect of circTMEM181 on soluble CD73 (sCD73), sCD73 was also detected in different conditions, but no significant differences of sCD73 were observed (Additional file 1: Fig. S7B and C).

To confirm how this CD39 (mostly expressed on macrophages)–CD73 (mostly expressed on HCC cells) axis affects the ATP–adenosine pathway in the tumor environment, we then monitored the level of eATP, AMP, and ADO in medium with different conditions. Only ample eATP, but not AMP or ADO, was detected in the medium of Li-7^OE or THP-1 cultured alone (Fig. 5c). Adding exosomes from Li-7^OE also could not elevate the ADO level in the medium when culturing THP-1 alone (Fig. 5c). However, when co-culturing THP-1 and Li-7^OE, the level of eATP decreased and AMP and ADO were increased in the medium (Fig. 5c). The increase in ADO and AMP in the co-culturing conditions could be largely inhibited by adding GW4869 or overexpressing miR-488-3p in THP-1 (THP-1^miROE) (Fig. 5c). We also investigated the function of TAM after co-culturing with Li-7^OE or CD39 inhibitor polyoxometalate 1 (POM1) (Fig. 5d). Interestingly, TAM showed a M2-like polarization characterized by high CD163 expression, more secretion of IL-10, and less secretion of TNF-α after co-culturing with Li-7^OE (Fig. 5d). These findings manifest that the interaction between HCC cells and THP-1 macrophages is involved in activation of the ATP–adenosine pathway and induces macrophage M2-like polarization.

We further tested the role of ADO in CD8⁺ T cell exhaustion. Co-cultured with ADO, CD8⁺ T cells showed high PD1 and TIM-3 expression and their activity was inhibited (Fig. 5E, F, and G). Similar results were observed in CD8⁺ T cells cultured with the supernatant from co-culture of macrophages and Li-7^OE (Sup.(Co)), but this inhibition could be rescued by culturing with the supernatant from co-culture of macrophages and Li-7^OE pre-adding POM1 (Sup.(Co + POM1)) (Fig. 5e, f, and g). However, when co-cultured with Li-7, only adding AMP, but not ADP or ATP, CD8⁺ T cells showed inhibited proliferation, indicating that HCC cells only degraded AMP into ADO by CD73 in the absence of CD39 expression (Fig. 5e and f). We then performed multiplex immunohistochemistry (mIHC) in HCC tissue to confirm CD39 and CD73 expression in the HCC microenvironment (Fig. 5h and Additional file 1: Fig. S8). CD39 was mostly expressed on endothelial cells, macrophages (CD68⁺), and some other immune cells (CD68⁻), but was not expressed on tumor cells (CK8⁺). As expected, CD73 was mostly expressed on CK8⁺ tumor cells, but barely expressed on macrophages (CD68⁺) (Fig. 5h and Additional file 1: Fig. S7D). Interestingly, CD39⁺ CD68⁺ macrophages were surrounded by CD73⁺ CK8⁺ HCC cells, indicating activation of the ATP–adenosine pathway by collaboration of tumors and macrophages in the HCC microenvironment (Fig. 5h).

The above data showed that HCC-cell-derived exosomal circTMEM181 enhanced CD39 expression in macrophages. In collaboration with CD73 on HCC, elevated CD39 degraded eATP into ADO, impairing the function of CD8⁺ T cells (Fig. 5i). To confirm these findings in vivo, we constructed subcutaneous tumor xenograft models in C57^wt mice and C57 mice with deleted CD39 (C57^{Entpd1−/−}). Moreover, Clodronate liposomes (CL) were also used to delete macrophages in the tumor microenvironment.

We found that the tumor volume of H22^OE was larger than that of H22^Ctrl in C57^wt mice (Fig. 6a). Meanwhile, no statistical difference in tumor volume was found between the CL group (H22^OE in C57 treated with CL) and the control group (H22^Ctrl in C57) (Fig. 6a). In C57^{Entpd1−/−}, both H22^OE and H22^Ctrl grew slower than that in the C57^wt model, but the tumor volume from H22^OE or H22^Ctrl showed no statistical difference (Fig. 6a). Tumors harvested from H22^OE in C57 mice showed the lowest infiltration of CD8⁺ T cells and the lowest granzyme B expression on CD8⁺ T cells among all the groups (Fig. 6b). Although deletion of CD39 in C57 mice did not increase the infiltration of CD8⁺ T cells compared with the control group (H22^Ctrl in C57), we found a significant increase in the granzyme B level on CD8⁺ T cells in the Entpd1^−/− groups (Fig. 6b).

Presently, several clinical trials targeting CD39 are ongoing (ClinicalTrials.gov identifiers: NCT03884556, NCT04261075, NCT04336098). Given the role of the cricTMEM181–macrophage ^CD39–HCC ^CD73 axis in tumor growth and CD8⁺ T cells, we then explored whether interfering with this axis, especially CD39, can enhance the efficiency of PD1 immunotherapy.

Subcutaneous tumor xenografts in C57 mice were constructed by injecting H22^OE or H22^Ctrl cells. Our findings showed a significantly larger tumor volume in the H22^OE group than in H22^Ctrl after anti-PD1 antibody therapy (Fig. 6c). In addition, H22^OE showed limited response to POM1 alone, while tumor growth was significantly inhibited when H22^OE were treated with a combination of anti-PD1 antibody and POM1 (Fig. 6c). Harvested tumors showed significant increases in CD8⁺ T cell infiltration and granzyme B expression on CD8⁺ T cells in the combined anti-PD1 antibody and POM1 group compared to the other two H22^OE groups treated with anti-PD1 antibody alone or with POM1 alone (Fig. 6d).

However, tumor size largely increased in the H22^Ctrl group after intravenous injection of exosomes enriched from H22^OE medium (Fig. 6e). A significant increase in CD39 expression on TAM (CD11b⁺ F4/80⁺) after infusion of exosomes enriched from H22^OE medium was found in harvested tumor. These CD39^high TAM tended to co-express CD163^high (Fig. 6f). In addition, significant decreases in CD8⁺ T cell infiltration and granzyme B expression on CD8⁺ T cells were found after infusion of exosomes enriched from H22^OE medium (Fig. 6f).

To further confirm whether specific depletion of CD39 on macrophages can enhance the effect of anti-PD1 antibody, C57-Lyz2^CreERT2 × C57-Entpd1^fl/fl mice were constructed (Additional file 1: Fig. S9A). Peripheral blood was harvested to verify selective deletion of CD39 on the macrophages after tamoxifen use (Additional file 1: Fig. S9B). H22^OE-labeled luciferase was then used to construct orthotopic liver xenografts in this model, and the mice were treated with anti-PD1 antibody. Interestingly, H22^OE showed poor response to anti-PD1 antibody, but selective deletion of CD39 on macrophages induced significant tumor shrinkage in response to anti-PD1 therapy (Fig. 6g). Flow cytometry analysis of CD45⁺ TIL showed that CD39^high was mostly expressed on CD11b⁺ myeloid cells and partially on CD8⁺ and CD4⁺ cells (Fig. 6h and i). Selective deletion of CD39 on macrophages effectively reduced CD39 expression on CD11b⁺ cells (including macrophages) in the tumor microenvironment (Fig. 6i). More CD8⁺ T cell infiltration and granzyme B expression on CD8⁺ T cells were found after selective deletion of CD39 in macrophages (Fig. 6j). Meanwhile, no significant change in CD73 expression was observed in H22 tumor cells after selective deletion of macrophage CD39 (Fig. 6k). Notably, we also found a significantly higher concentration of ADO and lower ATP in mouse serum upon selective deletion of CD39 on the macrophages (Fig. 6l).

Nine advanced HCC patients evaluated as PD or PR (iRECIST) after receiving the PD1 inhibitor nivolumab were analyzed. Five of the nine patients showed PD after PD1 immunotherapy, and four showed PR (Fig. 7a and b). The circTMEM181 level of the 9 tumor samples collected by puncture biopsy before immunotherapy together with the 6 tumor samples in our previous discovery cohort (Fig. 1a) was analyzed by qPCR. Our data confirmed a significant higher circTMEM181 expression in anti-PD1-resistant HCCs from the 15-patient cohort (Additional file 1: Fig. S10). Blood serum and PBMCs from the nine patients were also collected for further analysis of their immune status and blood exosomes. Seven major clusters were identified in the PBMCs: CD8 T (CD11b⁻, CD4⁻, CD8⁺), CD4 T (CD11b⁻, CD4⁺, CD8⁻), B cells (CD11b⁻, CD19⁺), NK (CD11b⁻, CD56⁺), cDC (CD11b⁺, CD141⁺), M1-like monocytes/macrocytes (CD11b⁺, CD163⁻), and M2-like monocytes/macrocytes (CD11b⁺, CD163⁺) (Fig. 7c). Although most clusters of immune cells remained relatively stable, patients evaluated as PD tended to have more CD11b⁺CD163⁺ M2-like monocytes/macrocytes and less CD11b⁺CD163⁻ M1-like monocytes/macrocytes in their PBMCs (Fig. 7d and e). We then explored the correlation between different immune cell clusters and the level of exosomal circTMEM181 in blood serum. Exosomal circTMEM181 showed a positive correlation tendency with cDCs, M2-like monocytes/macrocytes, and NK cells, but a negative correlation tendency with M1-like monocytes/macrocytes, CD8 T, CD4 T and B cells (Fig. 7f). We also found CD39 expression mostly on CD19⁺ B cells, CD11b⁺ myeloid cells, and some CD4⁺ T cells in the nine patients, and patients with PD expressed a higher level of CD39 on CD11b⁺ myeloid cells than patients with PR (Fig. 7g and h). Importantly, exosomal circTMEM181 in blood serum showed a positive correlation with CD39⁺ cells in PBMCs (Fig. 7i). Diagnostic liver biopsies of four patients before anti-PD1 therapy were analyzed by IHC. Our mIHC showed that CD39 was located on macrophages (CD68⁺), while CD73 was mostly located on HCC cells (CK8⁺) (Fig. 7j). More CD39⁺ macrophages (white arrow) were found in the tumor tissues from patients with PD (Fig. 7j), indicating an activated ATP–adenosine pathway in the tumor microenvironment of PD patients.